Kneedle is a general approach for detecting knee points in system behavior, applicable to a wide range of systems. A knee is defined as a point where the perceived cost to alter a system parameter is no longer worth the expected performance benefit. The authors define a knee using the mathematical concept of curvature for continuous functions, which measures how much a function differs from a straight line. This definition is application-independent and does not depend on the relationship between system parameters and performance. Kneedle is evaluated against existing algorithms on both synthetic and real data sets, showing its accuracy and performance in two different applications. The algorithm works by normalizing data, calculating the difference curve, and identifying local maxima in the difference curve, which indicate potential knee points. A threshold value is then used to determine if a knee has been detected. Kneedle can be used offline or online. In the online case, it can "correct" old knee values as new data points are received. The algorithm's online run time is bounded by O(n²). Kneedle is compared to other algorithms such as Angle-based, Menger, and EWMA. It is shown to have better accuracy and lower detection latency than these algorithms. Kneedle is applied to real-world systems, including a MapReduce-like system and a TCP-friendly network protocol. In the MapReduce-like system, Kneedle is used to detect knee points and reallocate tasks from slow nodes to fast nodes, reducing overall task completion time. In the network protocol, Kneedle is used to find the knee in the offered load versus packet delay curve, enabling effective congestion control. The authors conclude that Kneedle provides a general tool for detecting knee points in system behavior, which is useful for system designers who may not have the time to optimize specific systems. The methodology for evaluating knee detection approaches allows future designs to be compared in a straightforward manner.Kneedle is a general approach for detecting knee points in system behavior, applicable to a wide range of systems. A knee is defined as a point where the perceived cost to alter a system parameter is no longer worth the expected performance benefit. The authors define a knee using the mathematical concept of curvature for continuous functions, which measures how much a function differs from a straight line. This definition is application-independent and does not depend on the relationship between system parameters and performance. Kneedle is evaluated against existing algorithms on both synthetic and real data sets, showing its accuracy and performance in two different applications. The algorithm works by normalizing data, calculating the difference curve, and identifying local maxima in the difference curve, which indicate potential knee points. A threshold value is then used to determine if a knee has been detected. Kneedle can be used offline or online. In the online case, it can "correct" old knee values as new data points are received. The algorithm's online run time is bounded by O(n²). Kneedle is compared to other algorithms such as Angle-based, Menger, and EWMA. It is shown to have better accuracy and lower detection latency than these algorithms. Kneedle is applied to real-world systems, including a MapReduce-like system and a TCP-friendly network protocol. In the MapReduce-like system, Kneedle is used to detect knee points and reallocate tasks from slow nodes to fast nodes, reducing overall task completion time. In the network protocol, Kneedle is used to find the knee in the offered load versus packet delay curve, enabling effective congestion control. The authors conclude that Kneedle provides a general tool for detecting knee points in system behavior, which is useful for system designers who may not have the time to optimize specific systems. The methodology for evaluating knee detection approaches allows future designs to be compared in a straightforward manner.